Far from being as detailed as everyone above but just to leave some contribution to the 'Non-server' topic, here are some experiences that I've had.

Everything is already outdated, but it might be useful as some reference anyway.

I have a client running a pair of XCP-ngs 8.1 on Ryzen 7 1800X w/ Gigabyte AB350M-GAMING 3 motherboards, Kingston consumer grade SSDs (almost all the write disk activity goes to a NAS), doing HA with halizard nosan version (DRBD to synchronize storage) for years without a hitch. We have even moved the servers to another location, syncing them through an EoIP vpn with virtually no downtime.

And I have just assembled a Ryzen 5700G with an Asus B550M TUF Plus motherboard and XCP-ng 8.2.1 for a homelab. The new AM5 is still a bit expensive in Brazil for a homelab, and I didn't needed too much cpu power anyway, mostly memory and storage. I had an issue with the RTL8125B NIC (just now I noticed the very first message states the same) but adding an Intel NIC made everything perfectly fine.

@olivierlambert

Oh my, can't believe was that easy and I was fiddling around for hours with citrix links. In Brazil we would say "if this was a snake, you would be dead by now".

Thank you!

Just to give some closure that might help others, the issue was indeed realted to Networking.

I'm running the 8.2.1-test3 because I've started with a motherboard that has a RTL8125B NIC. Tried to switch to 8.3-alpha witht he same result, switched back to 8.2.1.-test3 and got an Intel 82574L NIC, now everything is working perfectly fine. The network speed of all VMs has actually increased significantly (this was another issue with the RTL8125B).